`sample` test case is broken with 1.75 about cargo-show-asm HOT 12 CLOSED

It is in original post in a .zip, together with .s for 1.73.
I am sorry for not providing one for 1.74, but this is what I had at hand.

from cargo-show-asm.

cargo run --release -- --manifest-path sample/Cargo.toml --native --intel sample::main --rust --no-default-features

works for me...

CI seem to detect it when it's there if I run with no argument... But fails to find it when it's called.

fn() -> () vs -> u32 shouldn't really affect anything, in both cases it should be a label like this: _ZN6sample4main17hecd2c58e95d09ed1E:.

I'll poke around, will let you know if I manage to reproduce it.

from cargo-show-asm.

Okay, got the jackpot. The return value was indeed not an issue.
The culprit lied in rust-lang/rust#116505
In short, with default features turned off, main was trivial enough to be marked as inline function automatically which then made the symbol weak. Since nobody was referencing it, it got stripped away.
Marking main in default-features=false config as #[inline(never)] or replacing 1 + 1 in it's body with a simple println!("foo") call (to make the main function sophisticated enough not to be subject to the new automatic marking as inline) makes the test pass again.

from cargo-show-asm.

It looks like main is now stripped? That seems somewhat plausible as it's defined in lib.rs, so who knows..

from cargo-show-asm.

Interestingly enough it works on my machine with 1.75.0

% cargo run --release --  --manifest-path sample/Cargo.toml --native --intel
    Finished release [optimized] target(s) in 0.02s
     Running `target/release/cargo-asm --manifest-path sample/Cargo.toml --native --intel`
    Finished release [optimized + debuginfo] target(s) in 0.00s

Try one of those by name or a sequence number
0 "<&T as core::fmt::Display>::fmt" [17]
1 "<sample::MyRngCore as rand_core::block::BlockRngCore>::generate" [18]
2 "core::ptr::drop_in_place<ahash::random_state::DefaultRandomSource>" [7]
3 "hashbrown::raw::RawTable<T,A>::reserve_rehash" [1238]
4 "hashbrown::raw::RawTable<T,A>::reserve_rehash::{{closure}}" [299]
5 "hashbrown::set::HashSet<T,S,A>::insert" [382]
6 "hashbrown::set::HashSet<T>::new" [241]
7 "make_bar" [8]
8 "sample::main" [751]
% rustc --version
rustc 1.75.0 (82e1608df 2023-12-21)

Can you try passing --everything to check if sample::main is there under some other name? Or any ideas how to reproduce it on my end?

from cargo-show-asm.

Stripping shouldn't affect anything - c-s-a doesn't run on compiled code, it asks rustc to dump generated .s file.

from cargo-show-asm.

Can you try building it with --no-default-features on sample? E.g:
cargo run --release -- --manifest-path sample/Cargo.toml --native --intel sample::main --rust --no-default-features
It looks like that makes a difference, as we also test the same scenario with default features on sample in CI and that passes even on 1.75.

By stripping I've meant dead code elimination, maybe there's now a MIR optimization that removes main from libs (so that it won't even have code generated for it) - though I don't see why it would only happen with default features turned off.
I'll post results with --everything in the morning.

from cargo-show-asm.

I have a hunch, will try that in the morning though;
With default features turned off, sample::main has a signature of fn() -> u32 which IIRC isn't a valid function signature for an entry point. With default features turned on, its signature is fn() -> () and that is a valid entry point signature. Maybe it is being preserved with default features (as it's an entry point) and being stripped away without any default features (as it's dead code?)

from cargo-show-asm.

Well, IMHO the return value should make a difference as to whether the code is stripped altogether or not.

from cargo-show-asm.

Can you attach the raw .s file?

from cargo-show-asm.

Good catch!

from cargo-show-asm.

Closing as it's fixed with e7e31ac71fdc8d350226a39ac2be5a4a47c6bac0

from cargo-show-asm.